-
Notifications
You must be signed in to change notification settings - Fork 4
/
reward_sim.py
282 lines (217 loc) · 11.9 KB
/
reward_sim.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
import math
import numpy as np
from shapely.geometry import Polygon
def reward_sim(X_reward, car_id, action_id, params):
episode = params.episode
complete_flag = params.complete_flag
l_car = params.l_car
w_car = params.w_car
w_lane = params.w_lane
num_lanes = params.num_lanes
l_road = params.l_road
num_cars = params.num_cars
# Off road penalty
Off_road = 0
Off_road_Penalty = -1e6
l_car_safe = 1.2*l_car # 1.2
w_car_safe = 1.2*w_car
Ego_rectangle = Polygon(
[[X_reward[0,car_id]-l_car_safe/2*math.cos(X_reward[2,car_id])+w_car_safe/2*math.sin(X_reward[2,car_id]), X_reward[1,car_id]-l_car_safe/2*math.sin(X_reward[2,car_id])-w_car_safe/2*math.cos(X_reward[2,car_id])],
[X_reward[0,car_id]-l_car_safe/2*math.cos(X_reward[2,car_id])-w_car_safe/2*math.sin(X_reward[2,car_id]), X_reward[1,car_id]-l_car_safe/2*math.sin(X_reward[2,car_id])+w_car_safe/2*math.cos(X_reward[2,car_id])],
[X_reward[0,car_id]+l_car_safe/2*math.cos(X_reward[2,car_id])-w_car_safe/2*math.sin(X_reward[2,car_id]), X_reward[1,car_id]+l_car_safe/2*math.sin(X_reward[2,car_id])+w_car_safe/2*math.cos(X_reward[2,car_id])],
[X_reward[0,car_id]+l_car_safe/2*math.cos(X_reward[2,car_id])+w_car_safe/2*math.sin(X_reward[2,car_id]), X_reward[1,car_id]+l_car_safe/2*math.sin(X_reward[2,car_id])-w_car_safe/2*math.cos(X_reward[2,car_id])]])
Upper_RoadBound_rectangle = Polygon(
[[0, w_lane*num_lanes],
[0, w_lane*num_lanes*2],
[l_road, w_lane*num_lanes*2],
[l_road, w_lane*num_lanes]])
Lower_RoadBound_rectangle = Polygon(
[[0, 0],
[0, -w_lane*num_lanes*2],
[l_road, -w_lane*num_lanes*2],
[l_road, 0]])
if (Ego_rectangle.intersects(Upper_RoadBound_rectangle) or
Ego_rectangle.intersects(Lower_RoadBound_rectangle)):
Off_road = Off_road + Off_road_Penalty
# Collision penalty
Colli = 0
Colli_Penalty = -1e6
l_car_safe = 1.1*l_car # 1.1
w_car_safe = 1.1*w_car
for id in range(0, len(X_reward[0,:])):
if id!=car_id:
Other_rectangle = Polygon(
[[X_reward[0,id]-l_car_safe/2*math.cos(X_reward[2,id])+w_car_safe/2*math.sin(X_reward[2,id]), X_reward[1,id]-l_car_safe/2*math.sin(X_reward[2,id])-w_car_safe/2*math.cos(X_reward[2,id])],
[X_reward[0,id]-l_car_safe/2*math.cos(X_reward[2,id])-w_car_safe/2*math.sin(X_reward[2,id]), X_reward[1,id]-l_car_safe/2*math.sin(X_reward[2,id])+w_car_safe/2*math.cos(X_reward[2,id])],
[X_reward[0,id]+l_car_safe/2*math.cos(X_reward[2,id])-w_car_safe/2*math.sin(X_reward[2,id]), X_reward[1,id]+l_car_safe/2*math.sin(X_reward[2,id])+w_car_safe/2*math.cos(X_reward[2,id])],
[X_reward[0,id]+l_car_safe/2*math.cos(X_reward[2,id])+w_car_safe/2*math.sin(X_reward[2,id]), X_reward[1,id]+l_car_safe/2*math.sin(X_reward[2,id])-w_car_safe/2*math.cos(X_reward[2,id])]])
if Ego_rectangle.intersects(Other_rectangle):
Colli = Colli + Colli_Penalty
# Safe zone violation penalty
Safe = 0
Safe_Penalty = -1e5 # 500
l_car_safe_front = 2*l_car
l_car_safe_back = 1.2*l_car
w_car_safe = 1.2*w_car
Ego_rectangle = Polygon(
[[X_reward[0,car_id]-l_car_safe_back/2*math.cos(X_reward[2,car_id])+w_car_safe/2*math.sin(X_reward[2,car_id]),X_reward[1,car_id]-l_car_safe_back/2*math.sin(X_reward[2,car_id])-w_car_safe/2*math.cos(X_reward[2,car_id])],
[X_reward[0,car_id]-l_car_safe_back/2*math.cos(X_reward[2,car_id])-w_car_safe/2*math.sin(X_reward[2,car_id]), X_reward[1,car_id]-l_car_safe_back/2*math.sin(X_reward[2,car_id])+w_car_safe/2*math.cos(X_reward[2,car_id])],
[X_reward[0,car_id]+l_car_safe_front/2*math.cos(X_reward[2,car_id])-w_car_safe/2*math.sin(X_reward[2,car_id]), X_reward[1,car_id]+l_car_safe_front/2*math.sin(X_reward[2,car_id])+w_car_safe/2*math.cos(X_reward[2,car_id])],
[X_reward[0,car_id]+l_car_safe_front/2*math.cos(X_reward[2,car_id])+w_car_safe/2*math.sin(X_reward[2,car_id]), X_reward[1,car_id]+l_car_safe_front/2*math.sin(X_reward[2,car_id])-w_car_safe/2*math.cos(X_reward[2,car_id])]])
for id in range(0,len(X_reward[1,:])):
if id!=car_id:
Other_rectangle = Polygon(
[[X_reward[0,id]-l_car_safe_back/2*math.cos(X_reward[2,id])+w_car_safe/2*math.sin(X_reward[2,id]), X_reward[1,id]-l_car_safe_back/2*math.sin(X_reward[2,id])-w_car_safe/2*math.cos(X_reward[2,id])],
[X_reward[0,id]-l_car_safe_back/2*math.cos(X_reward[2,id])-w_car_safe/2*math.sin(X_reward[2,id]), X_reward[1,id]-l_car_safe_back/2*math.sin(X_reward[2,id])+w_car_safe/2*math.cos(X_reward[2,id])],
[X_reward[0,id]+l_car_safe_front/2*math.cos(X_reward[2,id])-w_car_safe/2*math.sin(X_reward[2,id]), X_reward[1,id]+l_car_safe_front/2*math.sin(X_reward[2,id])+w_car_safe/2*math.cos(X_reward[2,id])],
[X_reward[0,id]+l_car_safe_front/2*math.cos(X_reward[2,id])+w_car_safe/2*math.sin(X_reward[2,id]), X_reward[1,id]+l_car_safe_front/2*math.sin(X_reward[2,id])-w_car_safe/2*math.cos(X_reward[2,id])]])
if Ego_rectangle.intersects(Other_rectangle):
Safe = Safe + Safe_Penalty
# Lane overlap violation penalty
# Adding when car's angle is 0, give the penalty.
eps = w_car/2.5
Lane_1 = Polygon([[0, w_lane-eps],[0, w_lane+eps],[l_road, w_lane+eps],[l_road, w_lane-eps]])
Lane_2 = Polygon([[0, 2*w_lane-eps],[0, 2*w_lane+eps],[l_road, 2*w_lane+eps],[l_road, 2*w_lane-eps]])
Ego_rectangle = Polygon(
[[X_reward[0,car_id]-l_car/2*math.cos(X_reward[2,car_id])+w_car/2*math.sin(X_reward[2,car_id]),X_reward[1,car_id]-l_car/2*math.sin(X_reward[2,car_id])-w_car/2*math.cos(X_reward[2,car_id])],
[X_reward[0,car_id]-l_car/2*math.cos(X_reward[2,car_id])-w_car/2*math.sin(X_reward[2,car_id]), X_reward[1,car_id]-l_car/2*math.sin(X_reward[2,car_id])+w_car/2*math.cos(X_reward[2,car_id])],
[X_reward[0,car_id]+l_car/2*math.cos(X_reward[2,car_id])-w_car/2*math.sin(X_reward[2,car_id]), X_reward[1,car_id]+l_car/2*math.sin(X_reward[2,car_id])+w_car/2*math.cos(X_reward[2,car_id])],
[X_reward[0,car_id]+l_car/2*math.cos(X_reward[2,car_id])+w_car/2*math.sin(X_reward[2,car_id]), X_reward[1,car_id]+l_car/2*math.sin(X_reward[2,car_id])-w_car/2*math.cos(X_reward[2,car_id])]])
Lane_overlap = 0
Lane_overlap_penalty = -1e4
if abs(X_reward[2, car_id]- X_reward[7,car_id] ) <= 1e-2:
if Ego_rectangle.intersects(Lane_1) or Ego_rectangle.intersects(Lane_2):
Lane_overlap = Lane_overlap + Lane_overlap_penalty
# Lane off-center penalty
Lane = 0
LC_penalty = -1e-2/(w_lane/2)
for i in range(0, num_lanes):
if X_reward[1,car_id] <= (i+1)*w_lane and X_reward[1,car_id] > (i)*w_lane:
lane = i+1
break
elif X_reward[1,car_id] > num_lanes*w_lane:
lane = num_lanes
elif X_reward[1,car_id] < 0:
lane = 1
lane_center = (lane-1)*w_lane + (w_lane/2)
Lane = Lane + LC_penalty*abs(lane_center-(X_reward[1,car_id]))
# Completion reward
Complete = 0
Complete_Penalty = -1
if X_reward[0,car_id]<X_reward[5,car_id]:
Complete = Complete + 1e-0 * Complete_Penalty/X_reward[5,car_id]*abs(X_reward[5,car_id]-X_reward[0,car_id])
Complete = Complete + 1e2 * Complete_Penalty/X_reward[6,car_id]*abs(X_reward[6,car_id]-(X_reward[1,car_id]))
Complete = Complete + 1e-4 * Complete_Penalty*abs(math.sin(X_reward[2,car_id] - X_reward[7,car_id] ))
else:
complete_flag[episode,car_id] = 1
# Speed reward
Speed = -1e-2*1/(X_reward[3, car_id])
# Effort penalty
Effort = 0
#Effort_penalty = -1e1
if(action_id==1 or action_id==2):
Effort = 0
else:
Effort = 0
# Local Reward
R_l = (Off_road + Colli + Safe + Complete + Speed + Effort + Lane + Lane_overlap)
params.complete_flag = complete_flag
### Social Reward
if X_reward[4, car_id]==1:
# Speed
v_target = params.v_target
v_sum = 0
for i in range(0, num_cars):
if X_reward[4, i] == 1:
v_sum = v_sum + X_reward[3, i]
v_avg = v_sum / params.num_AV
Vavg_penalty = -1e-1
R_Vavg = Vavg_penalty*abs(v_avg-v_target)
#Headway
num_AV_lane_1 = 0
num_AV_lane_2 = 0
num_AV_lane_3 = 0
AV_x_lane_1 = np.empty(shape=[1, 0])
AV_x_lane_2 = np.empty(shape=[1, 0])
AV_x_lane_3 = np.empty(shape=[1, 0])
for i in range(0, num_cars):
if X_reward[4, i] == 1:
if (X_reward[1, i] >= 0) and (X_reward[1, i] <= w_lane):
num_AV_lane_1 = num_AV_lane_1 + 1
#if X_reward[2,i]==0:
AV_x_lane_1 = np.append(AV_x_lane_1, X_reward[0, i])
elif (X_reward[1, i] >= w_lane) and (X_reward[1, i] <= 2*w_lane):
num_AV_lane_2 = num_AV_lane_2 + 1
#if X_reward[2,i]==0:
AV_x_lane_2 = np.append(AV_x_lane_2, X_reward[0, i])
elif (X_reward[1, i] >= 2*w_lane) and (X_reward[1, i] <= 3*w_lane):
num_AV_lane_3 = num_AV_lane_3 + 1
#if X_reward[2,i]==0:
AV_x_lane_3 = np.append(AV_x_lane_3, X_reward[0, i])
h_opt = l_car_safe
R_s = 0
Prop_headway_penalty = -1e-4
Head_way_Penalty = -1e2
Head_way_Penalty_tilt = -1e4*0
if num_AV_lane_1 ==0:
head_1 = 0
R_s = R_s + 0
elif num_AV_lane_1 ==1:
head_1 = 0
R_s = R_s + Head_way_Penalty
else:
head_1 = 0
head_sum_opt_1 = (num_AV_lane_1 - 1)*h_opt # desired value
if len(AV_x_lane_1)>=2:
for j in range(0, num_AV_lane_1-1):
head_temp = AV_x_lane_1[j+1] - AV_x_lane_1[j] - l_car
head_1 = abs(head_temp) + head_1
# if head_1 >= head_sum_opt_1:
R_s = R_s + abs(head_sum_opt_1 - head_1)*Prop_headway_penalty
# else:
# R_s = R_s + -1e6
else:
R_s = R_s + Head_way_Penalty_tilt
if num_AV_lane_2 ==0:
head_2 = 0
R_s = R_s + 0
elif num_AV_lane_2 ==1:
head_2 = 0
R_s = R_s + Head_way_Penalty
else:
head_2 = 0
head_sum_opt_2 = (num_AV_lane_2-1)*h_opt
if len(AV_x_lane_2)>=2:
for j in range(0, num_AV_lane_2-1):
head_temp = AV_x_lane_2[j+1] - AV_x_lane_2[j]-l_car
head_2 = abs(head_temp) + head_2
# if head_2 >= head_sum_opt_2:
R_s = R_s + abs(head_sum_opt_2 - head_2) * Prop_headway_penalty
# else:
# R_s = R_s + -1e6
else:
R_s = R_s + Head_way_Penalty_tilt
if num_AV_lane_3 == 0:
head_3 = 0
R_s = R_s + 0
elif num_AV_lane_3 == 1:
head_3 = 0
R_s = R_s + Head_way_Penalty
else:
head_3 = 0
head_sum_opt_3 = (num_AV_lane_3-1)*h_opt
if len(AV_x_lane_3)>=2:
for j in range(0, num_AV_lane_3-1):
head_temp = AV_x_lane_3[j+1]-AV_x_lane_3[j]-l_car
head_3 = abs(head_temp) + head_3
# if head_3 >= head_sum_opt_3:
R_s = R_s + abs(head_sum_opt_3 - head_3)*Prop_headway_penalty
# else:
# R_s = R_s + -1e6
else:
R_s = R_s + Head_way_Penalty_tilt
R_s = R_s + R_Vavg
else:
R_s = 0
R = R_l + R_s
return R, params